On 05/30/2015 06:59 PM, Xiao Guangrong wrote:
> Currently guest MTRR is completely prohibited if cache snoop is supported on
> IOMMU (!noncoherent_dma) and host does the emulation based on the knowledge
> from host side, however, host side is not the good point to know
> what the purpose of guest is. A good example is that pass-throughed VGA
> frame buffer is not always UC as host expected
>
> Besides that, there is a bugs in current code if noncoherent_dma is detected:
> KVM still uses hugepage even if the 4K pages in that ranges span between
> on different memory types and uses the cache type of first page to do memory
> mapping
>
> This patchset enables full MTRR virtualization and currently only works on
> Intel EPT architecture
>
> There is the code will be attached in this thread is used to check the effect
> of this patchset:
> [12834.843439] __INIT__.
> [12835.909109] Running in preload MTRR_TYPE_UNCACHABLE 4753900 ns.
> [12835.915232] Running in MTRR_TYPE_UNCACHABLE 5257522 ns.
> [12838.083948] Running in preload MTRR_TYPE_WRBACK 13266 ns.
> [12838.084998] Running in MTRR_TYPE_WRBACK 13099 ns.
>

Has been attached in this mail.